sign in sign up
<<
Home , Phospholipase D, STK11

ARTICLE

Received 9 Feb 2016 | Accepted 25 Apr 2017 | Published 26 Jun 2017 DOI: 10.1038/ncomms15747 OPEN -binding and of LKB1 by phosphatidic is essential for development and tumour suppression

Giada Dogliotti1,*, Lars Kullmann1,2,*, Pratibha Dhumale3,4, Christian Thiele1, Olga Panichkina1, Gudrun Mendl1, Roland Houben5, Sebastian Haferkamp6, Andreas W. Pu¨schel3,4 & Michael P. Krahn1,2

The serine/threonine LKB1 regulates various cellular processes such as proliferation, energy and cell polarity and is frequently downregulated in various tumours. Many downstream pathways controlled by LKB1 have been described but little is known about the upstream regulatory mechanisms. Here we show that targeting of the kinase to the membrane by a direct binding of LKB1 to is essential to fully activate its kinase activity. Consequently, LKB1 mutants that are deficient for membrane binding fail to activate the downstream target AMPK to control mTOR signalling. Furthermore, the in vivo function of LKB1 during development of Drosophila depends on its capacity to associate with . Strikingly, we find LKB1 to be downregulated in malignant melanoma, which exhibit aberrant activation of Akt and overexpress phosphatidic acid generating D. These results provide evidence for a fundamental mechanism of LKB1 activation and its implication in vivo and during .

1 Molecular and Cellular Anatomy, University of Regensburg, Universita¨tsstr. 31, 93053 Regensburg, Germany. 2 Internal Medicine D, University Hospital of Mu¨nster, Domagkstr. 3, 48149 Mu¨nster, Germany. 3 Institute for Molecular Cell Biology, University of Mu¨nster, Schlossplatz 5, 48149 Mu¨nster, Germany. 4 Cells-in-Motion Cluster of Excellence, University of Mu¨nster, D-48149 Mu¨nster, Germany. 5 Department of Dermatology, Venereology and Allergology, University Hospital Wu¨rzburg, Josef-Schneider-Str. 2, 97080 Wu¨rzburg, Germany. 6 Institute of Dermatology, University Hospital Regensburg, Franz-Josef-Strauss-Allee 11, 93053 Regensburg, Germany. * These authors contributed equally to this work. Correspondence and requests for materials should be addressed to M.P.K. (email: [email protected]).

NATURE COMMUNICATIONS | 8:15747 | DOI: 10.1038/ncomms15747 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15747

he serine/threonine kinase LKB1 is ubiquitously expressed overlapping with the AJ- and TJ markers E-Cadherin and ZO-1 and highly conserved throughout evolution. It has been (Fig. 1b and ref. 27). Similar, endogenous LKB1 localizes to the Tdemonstrated to function as a ‘master kinase’ potentially (lateral) membrane in epithelial cells in vivo (colon or salivary activating several downstream kinases1. Apart from its glands) (Fig. 1c,d). Remarkably, polarized epithelial cells in implication in carcinogenesis LKB1 plays a role in various culture (MDCK) and in situ (colon and salivary gland) do not cellular signalling pathways such as Wnt-, TGFb-signalling or exhibit nuclear staining of LKB1. the mTOR-pathway reviewed by Vaahtomeri and Makela2. In Drosophila cells endogenous LKB1 is also not found in the The latter one is controlled by LKB1-mediated activation of nucleus but localizes at the cortex of female germ line cells as well AMP-dependent kinase (AMPK), which is essential for cell as at the lateral membrane in epithelial cells22, whereas it shows a survival and polarity in Drosophila and vertebrates, in particular diffuse cytoplasmic pattern in neural stem cells of Drosophila under energetic stress3–7. Mechanistically, AMPK phosphorylates larval neuroblasts (NBs)28. Similar, the Caenorhabditis elegans (among others) Raptor (a core component of mTOR complex 1) orthologue, PAR-4, regulating asymmetric cell division of the and TSC2 (an mTOR inhibitor), resulting in reduced mTOR zygote, localizes to the entire cell cortex29. To test which activity8–15. Consequently, the LKB1-AMPK-axis is believed to mechanisms target LKB1 to the cortex, we raised an antibody be a key modulator in carcinogenesis and cell polarity16. against LKB1 and confirmed that in epithelial cells of the Apart from its function in and tumour embryonic epidermis, endogenous LKB1 is localized laterally, suppression, LKB1 is directly implicated in the establishment co-staining with a-spectrin (Fig. 1e), but also overlapping with and maintenance of cell polarity in different cell types and the zonula adherens (ZA), marked by Drosophila E-Cadherin organisms17. (DE-Cad) (Fig. 1e). Notably, in embryonic NBs, we detect a clear Although many downstream pathways mediating the function cortical LKB1 staining in interphase as well as during of LKB1 have been described, little is known about the upstream (Fig. 1f–h). However, in contrast to key regulators of asymmetric mechanisms regulating LKB1 activity. Two pseudokinases tightly cell division like the apical localized Bazooka (Baz) and control the localization and kinase activity of LKB1: STRADa the basally segregated adaptor protein Miranda (Mir), LKB1 does (Ste20-like kinase (Stlk) in Drosophila) and Mo25 form a stable not show a polarized distribution during mitosis (Fig. 1f–h). ternary complex with LKB1 (ref. 18). Both enhance the Thus, LKB1 localizes predominately to the (lateral) plasma export of LKB1 from the nucleus into the cytoplasm and increase membrane in polarized epithelial cells and NBs in Drosophila its kinase activity18,19. Secondly, the conserved C-terminus of and . LKB1 is farnesylated in vivo and thereby might directly interact with the plasma membrane to attach the protein to the cell cortex. Farnesylation of LKB1 is not essential for its localization. The Although the majority of the protein accumulates at the plasma C-terminus of LKB1 can be farnesylated21,30, establishing a membrane of polarized (epithelial) cells, farnesylation has been putative membrane targeting domain—however, this modifica- reported to be not essential for the (tumour suppressor) function tion does not seem to be important for its tumour suppressor of LKB1 in mammalian cells or mice but might be essential function in mammals20 and for viability of mice21. In the for oogenesis in Drosophila20–22. Therefore the question remains, Drosophila germ line, a block of farnesylation leads to a weaker whether membrane association of LKB1 is essential for the kinase cortical association and disturbed oocyte polarity22. To test activity and function of the protein during development and whether this is also true for epithelial cells and NBs, we used the tumour suppression. endogenous LKB1 promoter to express a GFP-tagged LKB1 wild- Here we report that binding of LKB1 to membranes by direct type protein or an LKB1 version with a Cys564-Ala substitution interaction with , in particular to phosphatidic acid, in the fly (farnesylation-deficient LKB1, lkb1::GFP-LKB1 ). is essential for its function in vivo during development of C564A Wild-type GFP-LKB1 is expressed at similar protein level as Drosophila. Membrane association of LKB1 is required for its endogenous LKB1 (Supplementary Fig. 1a) and localizes to the kinase activity and for efficient activation of AMPK in cultured lateral cortex of epithelial cells and the cortical membrane in NBs mammalian cells, thus contributing to the tumour suppressor indistinguishable from the endogenous protein (Fig. 1i,k). function of LKB1. Strikingly, we reveal a strong correlation between Surprisingly, in the embryonic epidermis and embryonic NBs, overexpression of Phospholipase D (PLD), which increases cellular GFP-LKB1 shows the same as its levels of phosphatidic acid, downregulation of LKB1 and enhanced C564A wild-type counterpart (Fig. 1j,l). The farnesylation-deficient activity of mTOR in malignant melanoma, thus likely contributing protein shows a more cytosolic distribution only in epithelial to the of malignant melanoma. cells surrounding the oocyte (follicular epithelium) although a substantial fraction of the protein is still associated with the Results lateral membrane (Fig. 1n compared to wild-type GFP-LKB1 in Fig. 1m). LKB1 localizes to the cortex of epithelial cells and neuroblasts. Notably, LKB1 expressed from its endogenous promoter The activity of can be modulated for instance by directly C564A is able to rescue an lkb1-null allele (lkb1X5) to a large extent influencing their enzymatic activity (e.g., by a conformational (52% surviving flies in comparison to 69% for wild-type LKB1, change via phosphorylation) or by targeting the protein to Fig. 1o), indicating that farnesylation of LKB1 is not essential for different subcellular compartments. In cultured mammalian cells, the function of the protein in vivo. LKB1 accumulates in the nucleus in many cell lines and only a minor fraction of the protein is found in the cytoplasm or at the cytocortex23–25. LKB1 has been demonstrated to shuttle from the LKB1 directly binds to phospholipids. In order to further cytoplasm into the nucleus and back, with its two co-factors, elucidate the targeting of LKB1 to the plasma membrane, we used STRADa and Mo25 enhancing cytoplasmic localization and Schneider R þ (SR þ ) cells as they do not exhibit an intrinsic activating the kinase18,19,26. polarity and do not express transmembrane proteins like DE-Cad, Whereas endogenous LKB1 accumulates predominately in Crumbs or Echinoid, qualifying them as a model for the analysis the nucleus of non-transformed fibroblasts (Fig. 1a), epithelial of direct plasma membrane targeting. As expected, GFP-LKB1 cells (Madin Darby Canine Kidney, MDCK) exhibit a staining of localizes to the plasma membrane in transfected S2R þ cells endogenous LKB1 exclusively at the cell-cell contacts, partly (Fig. 2c). In contrast, the farnesylation motif alone (GFP fused to

2 NATURE COMMUNICATIONS | 8:15747 | DOI: 10.1038/ncomms15747 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15747 ARTICLE

a IMR90 i

LKB1 DAPI Merge Merge b MDCK GFP-LKB1 DE-Cad Dlg +DAPI j

Merge Merge GFP-LKB1C564A DE-Cad Dlg +DAPI LKB1 E-Cad ZO-1 +DAPI k cdColon Salivary gland

Merge GFP-LKB1 Baz Mir +DAPI LKB1 LKB1 e l

Merge DE-Cad LKB1 α-spectrin +DAPI Merge

f GFP-LKB1C564A Baz Dlg +DAPI m Merge GFP-LKB1 DE-Cad Dlg +DAPI Merge Baz LKB1 Mir +DAPI n g Merge GFP-LKB1C564A DE-Cad Dlg +DAPI

o % Lethality/survival 100 * Merge 90 * Baz LKB1 Mir +DAPI *** 80 * h 70 60 50 40 Merge 30 Baz LKB1 Mir +DAPI 20 10 0 Embryonic lethal L1 & L2 lethal L3 lethal Pupal lethal Survivors ×5 ×5 Ikb 1 Ikb1::GFP-LKB1ΔLB C564A; lkb1 ×5 Ikb1::GFP-LKB1; lkb1 ×5 Ikb1::GFP-LKB1ΔLB C564A-PH(PLCδ); lkb1 ×5 Ikb1::GFP-LKB1 ; lkb1 ×5 C564A Ikb1::GFP-LKB1ΔLB C564A-PH(Akt); lkb1 ×5 Ikb1::GFP-LKB1ΔLB; lkb1

Figure 1 | LKB1 localizes to the plasma membrane in polarized epithelial cells and neuroblasts. (a) In non-transformed mammalian fibroblasts (IMR90 cells), endogenous LKB1 accumulates mostly in the nucleus. (b) In polarized epithelial cells (MDCK), LKB1 is targeted to the cell-cell contacts, partly colocalizing with E-Cadherin (E-Cad) and Zonula occludens protein 1 (ZO-1). (c,d) Sections of paraffin-embedded colon (c) and salivary gland tissues (d) show a localization of LKB1 at the lateral plasma membrane in situ (arrows). (e–h) LKB1 localizes to the lateral plasma membrane in epithelial cells of the

Drosophila embryonic epidermis (e) and to the cortex of neuroblasts (f–h). (i–n) Wild-type GFP-LKB1 as well as farnesylation deficient LKB1 (LKB1C564A) expressed from its endogenous promoter localize correctly to the (lateral) membrane of epithelial cells of the embryonic epidermis (i,j), of the follicular epithelium (m,n) and of neuroblasts (k,l). (o) Lethality tests of LKB1 variants as described in the methods section. Scale bars are 20 mmina–d,5mm in e–n. Experiments were performed in triplicates. Error bars represent s.d. and statistical significance was determined using ANOVA: Po0.0001, ***Po0.01, *P40.05, not significant (NS).

the last 15 amino of LKB1, GFP-LKB1552-C) is not sufficient Several protein domains are known to facilitate protein– to target the protein to the cortex (Fig. 2d), which is in line interactions. Beside larger domains, polybasic motifs have with the hypothesis that stable membrane binding requires been described for several proteins to bind to (phospho)-, protein modifications (e.g., ) or membrane binding in particular to phosphatidic acid32. Indeed, a C-terminal domains31 in addition to conjugation with farnesyl acid. In polybasic motif (aa 539–551, termed lipid-binding motif ¼ LB, contrast, a longer C-terminal fragment of LKB1 (LKB1512-C and Fig. 2a) facilitates a direct binding of DmLKB1 to PA, PtdIns(5)P, LKB1536-C) exhibits a robust cortical localization apart from a PtdIns(3,4,5)P3 and PtdIns(4,5)P2 as determined by a lipid nuclear staining (Fig. 2e; Supplementary Fig. 2a). Furthermore, overlay assay (Fig. 2b). A strong binding to PA (and to some in embryos, a substantial fraction of GFP-LKB1512-C localizes to extent to PtdIns(3,4,5)P3 and PtdIns(4,5)P2) was confirmed in the lateral membrane (Fig. 2g). liposome flotation assays (Supplementary Fig. 2b). Membrane association can be achieved not only by the of this motif (mutation of Arg and Lys in LKB1-LB to interaction with transmembrane or membrane associated Ala, LKB1DLB) abolished liposome association (Supple- proteins but also by the direct binding to the . mentary Fig. 2b) as well as membrane association in cultured cells

NATURE COMMUNICATIONS | 8:15747 | DOI: 10.1038/ncomms15747 | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15747

LKB1 abLKB1 ΔLB VGKKK SALKRRAKKLTSCISVRKLSHCRTS Lysophosphatic acid Sphingosine-1phosphate 538 567 Lysophosphocholine Ptdlns(3,4)P2 NH Kinase domain COOH 2 Ptdlns Ptdlns(3,5)P2 1 567 Ptdlns(3)P Ptdlns(4,5)P2 Ptdlns(4)P Ptdlns(3,4,5)P3 Ptdlns(5)P Phosphatidic acid (PA) (PE) Phosphatidylserine (PS) (PC) Blank

c def h Merge GFP-LKB1 DE-Cad Dig +DAPI i GFP-LKB1 Merge GFP-LKB1 GFP-LKB1552-C GFP-LKB1512-C ΔLB GFP-LKB1ΔLB DE-Cad Dig +DAPI g j Merge GFP-LKB1512-C DE-Cad Dig +DAPI

Merge GFP-LKB1ΔLB Baz Mir +DAPI k

ΔLB GFP-LKB1 Dig DE-Cad Merge+DAPI

Figure 2 | LKB1 is recruited to the lateral plasma membrane by direct binding to phospholipids. (a) Schematic drawing showing Drosophila LKB1 with basic residues and the farnesylation motif within its C-terminal region highlighted. (b) Lipid overlay assay of recombinant LKB1 and LKB1DLB ( ¼ LKB1K546A R547A R548A K550A K551A K539A K540A K541A) using diverse lipids spotted on nitrocellulose membrane. (c–f) GFP-LKB1 and GFP-LKB1512-C localize to the cell cortex of transfected S2R cells (c,e), whereas a fusion protein of the last 16aa of LKB1 with GFP (GFP-LKB1552-C) or GFP-LKB1DLB does not (d,f). (g–j) GFP-LKB1512-C accumulates at the lateral membrane of epithelial cells in the embryonic epidermis (g), similar to full length GFP-LKB1 (h), whereas GFP-LKB1DLB exhibits a cytosolic localization in epithelial cells of the embryonic epidermis (i) and of the follicular epithelium (k) as well as in neuroblasts (j). Scale bars are 5 mm.

(Fig. 2f) and in Drosophila epithelia and neural stem cells indicating that the association of LKB1 with membranes is (Fig. 2h–k). essential for protein stability. In contrast, transient membrane association of LKB1DLB, which is probably mediated by the farnesylation anchor, is sufficient for protein stabilization, as no Membrane-bound LKB1 is crucial for Drosophila development. differences in protein levels are detectable between wild-type LKB1 has been implicated in various cellular functions, among LKB1 and lipid-binding-deficient LKB1 (Fig. 3d). them cell proliferation control, suppression of tumour growth However, overexpression of GFP-DmLKB1DLB C564A in and regulation of cell polarity in various cell types2,17. To address lkb1-mutant flies using the UAS/GAL4 system instead of the the question whether membrane targeting of LKB1 is crucial for endogenous lkb1-promoter does not result in detectable rescue its physiological function, we performed rescue experiments with capacity (Supplementary Fig. 3a). Similar, the overexpression GFP-LKB1 expressed from its endogenous promoter. As (rough eye formation) of LKB1 is abolished in indicated above, not only wild-type LKB1 but also LKB1C564A GFP-DmLKB1DLB C564A similar to a kinase dead version can rescue the embryonic lethality of maternal and zygotic (Fig. 3e). Thus, we assume that it is very unlikely that impaired mutant flies to a substantial extent, indicating that wild-type and functionality of GFP-DmLKB1DLB C564A is only due to the farnesylation-deficient GFP-LKB1 can substitute endogenous instability of the protein. LKB1 in all tissues (Fig. 1o). Notably, GFP-DmLKB1DLB still To further substantiate this assumption, we fused two different exhibits a residual rescue capacity (8% of surviving flies in heterologous membrane-binding domains to the C terminus contrast to 69% for wild-type DmLKB1, Fig. 1o), which might be of GFP-LKB1DLB C564A: The PH-domain of human phospholipase due to a transient or weak membrane binding mediated by the Cd (ref. 33) that preferentially binds to PtdIns(4,5)P2 (resulting in farnesylation anchor. Indeed, removal of the farnesylation motif GFP-LKB1DLB C564A-PH(PLCd)), but is also capable of binding 34 in GFP-LKB1DLB (GFP-LKB1DLB C564A) totally abolishes the to PA with a high affinity , and the PH-domain of human rescue capacity of the mutant protein and all mutant flies die Akt1, (GFP-LKB1DLB C564A-PH(Akt1)) with a high affinity to during embryonic stages like the null allele lkb1X5 (Fig. 1o). PtdIns(3,4,5)P3 (ref. 35). Similar to PLCd, Akt1 has recently 36 Moreover, lkb1-mutant flies expressing GFP-LKB1DLB C564A been demonstrated to bind PA in vitro . Strikingly, both exhibit the same polarity defects as the null allele alone (e.g., chimeric proteins stabilize the protein expression of GFP- disruption of the anterior–posterior polarity of ovaries, Fig. 3c, LKB1DLB C564A (Fig. 3d), localize at least partly to the lateral compared with wild-type LKB1 rescue in a and lkb1-mutant plasma membrane (Supplementary Fig. 4a,b) and are capable phenotype in b). of rescuing the lkb1 null allele to a large extent (56 and Western blot analyses of lysates from embryos expressing 48% respectively, Fig. 1o). These results confirm our hypothesis LKB1 variants from the endogenous promoter reveal that that membrane association of LKB1 is crucial for its function GFP-LKB1DLB C564A is unstable or rapidly degraded (Fig. 3d), in vivo.

4 NATURE COMMUNICATIONS | 8:15747 | DOI: 10.1038/ncomms15747 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15747 ARTICLE

a fgS2R+ STRADα + S2R+ STRADα +

+ PA

Merge + LB-C564A ×5 Δ KD GFP-LKB1; Ikb1 Gurken Staufen DAPI 1 1 1

LB-C564A LB-C564A ×5 + PA Δ Δ GFP LKB LKB LKB kDa Ikb1 1 1 1 1 b 32 130 P GFP-LKB1 GFP GFP + PA LKB LKB LKB LKB kDa 100 32 55 32 130 Merge + P GST- P GFP-LKB1 Gurken Staufen DAPI AMPK108–280 100

130 32 GFP-LKB1 P GST- 55 c 100 AMPK108–280 GFP- 70 130 STRADα GFP-LKB1 100 55 GFP-LKB1 Merge + GFP- ×5 LB C564A; Ikb1 Gurken Staufen DAPI STRADα 70 40 55 d Embryonic lysates 35 ) Δ GFP 40 35 CCB GST- 55 GFP AMPK108–280

LB LB C564A LB C564A PH(PLCLB C564A PH(Akt) CCB 55 Δ C564A Δ Δ Δ LKB auto-p 1 1 1 1 1 1 1 100 GST-AMPK 80 % 60 *** LKB1 auto-p 40 200 20 GFP-LKB GFP-LKB GFP-LKB GFP-LKB GFP-LKB GFP-LKB *** *** ** 0 150 GFP 100 AMPK-p % 100 100 NS 40 80 50 % 60 40 *** 0 20 *** *** 0 AMPK-p wt LKB1 LKB1KD LKB1ΔLB C564A 300 250 *** m m μ e μ 100 100 200 % 150 100 NS 50 0

100 μm100 μm

Figure 3 | Membrane-association of LKB1 is essential for its kinase activity and function in vivo. (a–c) Immunostainings of lkb1-mutant ovaries

(b) exhibit a disturbed localization of Staufen, which is rescued by expression of GFP-LKB1 (a) but not of GFP-LKB1DLB C564A (c). (d) Immunoblotting of LKB1 variants expressed in embryos demonstrating that the protein stability of GFP-LKB1DLB C564A is strongly reduced. (e) Overexpression of LKB1, but not of a kinase-dead version of LKB1 (LKB1D317A ¼ LKB1KD) or GFP-LKB1DLB C564A results in an impaired eye morphology (‘rough eye’ phenotype). (f,g) In vitro kinase assays of GFP-LKB1/GFP-STRAD purified from transfected S2R cells demonstrate a strong decrease in the kinase activity of LKB1DLB C564A measured 32 32 by autophosphorylation ( P GFP-LKB1) and phosphorylation of recombinant AMPK ( P GST-AMPK108-280)(f). The addition of PA-enriched liposomes (as described in materials and methods) increased the kinase activity of wild type but not of membrane-binding deficient LKB1 (g). Activity of GFP-LKB1 was set as 100%. Inputs were visualized by Western Blot against GFP (GFP, GFP-Stlk and GFP-LKB1) and by Coomassie Brilliant Blue staining (CCB, GST-AMPK108-280) Scale bars are 5 mminc–k and m, 100 mmino and p,10mminl and n. Experiments were performed in triplicates. Error bars represent s.d. and statistical significance was determined using ANOVA: Po0.001, ***Po0.01, **Po0.05, not significant (NS).

To test whether membrane association is essential for the The function of human LKB1 depends on membrane binding. kinase function of LKB1, we performed in vitro kinase assays As association with the membrane is essential for LKB1 to using the GFP-DmLKB1/STRADa complex purified from accomplish its function during development of Drosophila,we transfected S2R cells. Indeed GFP-DmLKB1DLB C564A exhibits a further investigated whether this mechanism is conserved strongly decreased kinase activity (46% in autophosphorylation throughout evolution and controls the activity of human LKB1 and 30% in AMPK-phosphorylation relative to wild-type (hLKB1). Similar to its fly homologue, mutation of the polybasic GFP-LKB1, Fig. 3f). Furthermore, addition of PA-enriched region at the C terminus of hLKB1 together with mutation of the liposomes to GFP-DmLKB1/STRADa strongly increases the farnesylation motif (Fig. 4a) resulted in an impaired membrane kinase activity of the LKB1 complex (autophosphorylation localization in HeLa cells (Fig. 4b,c). 1.6-fold and AMPK-phosphorylation 2.4-fold (Fig. 3g). Activation of AMPK is one of the most important functions of Reduced kinase activity of LKB1DLB C654A is neither due to LKB1 during tumour progression, which is further enhanced under impaired association with the canonical binding partners energetic stress. As activated AMPK directly and indirectly inhibits STRADa or Mo25, which enhance LKB1 activity, nor due to mTOR8–15, the LKB1-AMPK pathway controls mTOR activity, decreased binding (Supplementary Fig. 3b). Therefore, e.g., upon aberrant activation of Akt, which occurs in various types these data suggest that the lipid binding capacity of DmLKB1 is of tumours with loss of function of PTEN or PI3-Kinase gain of essential for its kinase function. function mutations37. Loss of membrane binding capacity of

NATURE COMMUNICATIONS | 8:15747 | DOI: 10.1038/ncomms15747 | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15747

a b HeLa Dm LKB1 Hs LKB1 GFP-hLKB1 DAPI Merge c HeLa

GFP-hLKB1ΔLB C430A DAPI Merge

d HeLa e HeLa

Y511F LB C430A KD Δ 1 1 1

Blot Control hLKB hLKB hLKB kDa + PLD2 + PLD2 Y511F 1 1 1 70 pAMPK kDa Blot PLD2 PLD2 Control hLKB hLKB hLKB 70 AMPK 70 pAMPK

pS6K 70 AMPK

S6K 70 pS6K 70 100 pMARK S6K 70

100 GFP pMARK (hLKB1) 70 GFP (hLKB1) 70 Actin 40 Actin 40

Cell viability (%) - 12 h AICAR 500 f pAMPK 400 100 ** ** ** 300 80 NS % 200 60 NS 100 40 0 20 0 1 LB 1 KD 1 Δ hLKB 1 C430A Control LB C430A hLKB hLKB 1 Δ hLKB hLKB

Figure 4 | Membrane binding of hLKB1 is essential for its function in mammalian cells. (a) Alignment of DmLKB1 and hLKB1 reveals several positively charged aa in the C terminus of hLKB1. Mutated residues are marked with bold lines. (b,c) GFP-hLKB1 localizes predominately to the nucleus and plasma membrane of HeLa cells cotransfected with STRADa (b), whereas GFP-hLKB1DLB C430A is not recruited to the cortex (c). (d,e) Immunoblottings of lysates from transfected HeLa cells, which lack substantial LKB1 expression, demonstrate an increase of activated AMPK (phospho-T172 AMPK), activated MARK (phospho-MARK) and decreased mTOR activation (measured by phosphorylation of S6K) upon GFP-hLKB1 þ STRADa transfection, which is not observed in GFP-hLKB1DLB C430A þ STRADa transfected cells (d). Co-transfection of PLD2 but not of a catalytically reduced variant (PLDY511F) together with hLKB1 results in a further increase of AMPK activation (e). The intensity of pAMPK bands was quantified (normalized against total AMPK) (e, lower panel, control was set as 100%). (f) Cell viability of HeLa cells under energetic stress (induced by incubation with AICAR for 12 h) was estimated using the MTT assay (as described in the methods section). Scale bars are 10 mm. Experiments were performed in triplicates. Error bars represent s.d. and statistical significance was determined using ANOVA: Po0.01, **Po0.05, not significant (NS). hLKB1 strongly decreases its ability to activate AMPK in HeLa cells AMPK phosphorylation in vitro, whereas Phosphatidylcholine (Fig. 4d), which lack substantial endogenous LKB1 expression, and (PC) alone or PtdIns(4,5)P2 and PtdIns(3,4,5)P3-enriched consequently leads to in metabolically stressed cells liposomes had only a small effect on the activity of hLKB1 (Supple- (Fig. 4f). Apart from AMPK activation, hLKB1DLB C430A fails to mentary Fig. 3c,d). activate two other kinases of the AMPK-family, MARK (Fig. 4d) and SadA (Supplementary Fig. 5a). Furthermore, hLKB1DLB C430A is no longer able to inhibit mTOR (as estimated by phospho- Overexpression of PLD2 enhances LKB1 activity. PA is gener- rylation of p70-S6-Kinase 1, pS6K, Fig. 4d). Vice versa, addition of ated (among other pathways) from PC by two isoforms of PLD, PA-enriched liposomes to a recombinant hLKB1/hSTRADa/ PLD1 and 2 (ref. 38). Expression of PLD2 but not of a variant of hMo25 complex enhanced autophosphorylation of hLKB1 and PLD2 with a reduced catalytic activity (PLD2 Y511F (ref. 39))

6 NATURE COMMUNICATIONS | 8:15747 | DOI: 10.1038/ncomms15747 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15747 ARTICLE

abc f 0 NS 1 100 **** >1 **** 80 **

60 Control hLKB1 hLKB1 C430A % 40 de 20

0

ΔLB hLKB1 C430A hLKB1Δ hLKB1Δ Control LB C430A LB LB C430A hLKB1 Δ hLKB1 hLKB1

Figure 5 | Induction of multiple axons in primary rat hippocampal depends on membrane binding of hLKB1. (a–f) Hippocampal neurons from E18 rat embryos were transfected at 0 d.i.v. with vectors for GFP (control), GFP-LKB1 or GFP fusion proteins for the indicated LKB1 mutants and STRAD and analysed at 3 d.i.v by staining with the Tau-1 antibody (axonal marker). Representative images of transfected neurons are shown. The scale bar is 20 mm. (f) The development of neuronal polarity was analysed by counting the number of neurons without an axon (0, black), with a single axon (1, gray) or with multiple axons (41, white). Most primary rat hippocampal neurons transfected with GFP alone (control) develop only a single axon (a, quantified in f).

In contrast, neurons overexpressing GFP-hLKB1 þ STRADa or GFP-hLKB1C430A þ STRADa (b,c,f) frequently establish two or more axons, whereas hLKB1 variants, which are deficient in membrane binding, fail to induce a multiple axon phenotype (d–f). Experiments were performed in triplicates. Error bars represent s.d. and statistical significance was determined using ANOVA: Po0.0001, ****Po0.01, **P40.05, not significant (NS).

together with hLKB1 results in a further increase of AMPK and mTOR activation48–50 only if in addition the expression of LKB1 MARK activation (Fig. 4e), suggesting that PLD2-induced is reduced, because PA-mediated LKB1/AMPK activation production of PA enhances LKB1 activity in vivo. Conversely, normally counteracts mTOR activity. To test this hypothesis, inhibition of PLD reduces LKB1’s capacity to activate AMPK we investigated whether increased expression of PLD, aberrant (Supplementary Fig. 5b). However, basal levels of activated AMPK activation of Akt and decreased LKB1 expression correlate are maintained, indicating that other (e.g., Diacylglycerol with enhanced mTOR activity in melanoma. Indeed, biopsies of kinase) can compensate loss of PLD activity. This is in line with melanoma primary tumours show a strong correlation between the observation that C. elegans, Drosophila and mouse mutants these four parameters. In 81% of all analysed tumours PLD2 lacking PLD enzymes do not exhibit as dramatic as expression as well as phospho-Akt and mTOR activity were LKB1DLB C564A (refs 40–42). elevated while the expression of LKB1 was decreased (Fig. 6a–e). Only 12% of the samples showed mTOR activation in the Membrane-binding-deficient LKB1 fails to induce multiple axons. presence of stimulating phospho-Akt and PLD2-overexpression Overexpression of LKB1 induces the extension of multiple although LKB1 staining appeared normal, which might be axons43–45. Consistently, overexpression of wild-type GFP-hLKB1 explained by inactivating of LKB1, which do not together with STRADa results in the formation of multiple axons affect protein expression or stability. In contrast, melanocytes in by the majority of transfected rat hippocampal neurons in culture biopsies of healthy skin exhibited a strong LKB1 expression, as (Fig. 5b,f). Remarkably, expression of farnesylation-deficient well as low level of PLD2, phospho-Akt and mTOR activation (Fig. 6a–d, arrows). Melanocytic nevi, a benign proliferation of hLKB1C430A induces this phenotype as efficiently as wild-type LKB1 (Fig. 5c,f). By contrast, mutation of the C-terminal basic melanocytes, which can give birth to malignant melanoma, show in many cases elevated PLD2 expression as well as region (hLKB1DLB) almost abolished the effect on axon formation (Fig. 5d,f). Inactivation of the farnesylation signal in lipid-binding- activation of Akt, whereas LKB1 expression is still high in most cases, suppressing aberrant mTOR activation, which occurred deficient hLKB1 (hLKB1DLB C430A) did result in further reduction of LKB1 activity in this assay (Fig. 5d,f). These results confirm the only in 25% of the analysed specimen (Fig. 6a’–d’). Thus importance of the C-terminal membrane for the in melanoma, loss of LKB1 in PLD2 overexpressing tumours function of LKB1 in mammalian cells. may further contribute to mTOR activation, whereas in benign neoplasia higher LKB1 level counterbalance aberrant PLD2 activation. Expression of LKB1 is downregulated in malignant melanoma. As we have demonstrated that overexpression of PLD2 in cell culture results in increased activity of LKB1 (reflected by enhanced Discussion AMPK activation, Fig. 4e), we next assessed the pathophysiological In this study, we have elucidated a conserved mechanism relevance of our findings regarding tumour formation in vivo. regulating LKB1 activity and function in vivo and during tumour Interestingly, expression of PLD2 is upregulated in several types of suppression (Fig. 7). We have further established the first , including melanoma, and high PLD2 expression correlates link between PLD-mediated production of PA and activation of with poor survival rates of patients46,47.Inculturedmelanomacells LKB1. Our data suggest that membrane targeting of LKB1 lacking detectable LKB1 expression (IGR37), membrane binding of by direct binding to PA is essential for the function of LKB1 LKB1 is essential for efficient AMPK (and MARK) activation, during Drosophila development and for activation of AMPK suppression of mTOR and cell survival under energetic stress and suppression of mTOR activity in cultured mammalian cells. (Supplementary Fig. 6). Furthermore, downregulation of LKB1 in malignant melanoma Our data suggest that an aberrant increase in PLD expression specimen, which exhibit activated Akt and overexpression of in cancer cells will lead to increased PA levels and subsequent PLD2, correlates with enhanced mTOR activity, indicating that

NATURE COMMUNICATIONS | 8:15747 | DOI: 10.1038/ncomms15747 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15747

e Healthy % Melanoma skin 90 (n = 67) 80 70 60 50 40 Neavus 30 20 10 0 hLKB1 hLKB1 Others PLD2 PLD2 pAkt pAkt pS6K pS6K Melanoma

hLKB1 PLD2 pAkt pS6K

Figure 6 | Downregulation of hLKB1 correlates with increased mTOR signalling in malignant melanoma overexpressing PLD2. Immunostainings of paraffin-embedded tissue sections of normal skin (a–d), neavi (a0–d0) and primary malignant melanoma tumours (a00–d00). (a) hLKB1 is highly expressed in melanocytes in the stratum basale of the epidermis in situ (a, arrows), and still detectable at substantial levels (a0) whereas it is almost undetectable in the majority of malignant melanoma (a00). (b) PLD2 is not detectable in melanocytes of healthy skin sections (b, arrows) but becomes overexpressed in naevi (b0) and primary tumours (b00). (c) Akt is activated (indicated by staining of phospho Akt, pAkt) in naevi (c0) and malignant melanoma (c00), whereas it is not detectable in melanocytes of healthy skin samples (c). (d) Melanocytes in healthy skin biopsies (d) show only faint nuclear staining for phospho S6K (pS6K), which was used as to evaluate mTOR activation. Naevi (d0) exhibit only very few pS6K positive cells, whereas malignant melanoma show strong cytoplasmic and nuclear staining of pS6K (d00). (e) Quantification of melanoma biopsies scored for the indicated correlations. ‘Others’ are tumours, which do not exhibit increased phospho-Akt or PLD2 staining.

Healthy cells Malignant melanoma

PLD PA PC PLD PIP3 PIP3 PA PC

P P Akt LKB1 P P Akt AMP LKB1 P P mTOR AMPK P P P P mTOR AMPK

Balanced cell prolifrration and cell growth tumour suppression Cell proliferation Cell growth Tumour progression

Figure 7 | Simplified model of LKB1 activation by PLD2-produced PA and its downstream effects on AMPK activation and mTOR inhibition. Under physiological conditions in non-tumourous cells, PA is produced by PLD (and other enzymes), resulting in an activation of mTOR and LKB1. Due to the inhibition of mTOR by LKB1-activated AMPK, the final result of increased PA level is a balanced cell proliferation and -growth. In contrast, in malignant melanoma, increased PI3K/Akt-signalling and enhanced production of PA by overexpression of PLD results in aberrant mTOR activation which is not counterbalanced by LKB1/AMPK and thus leads to increased cell proliferation and -growth and impaired tumour suppression. the mechanism described in this study contributes to pathogen- culture experiments, no difference was found between wild-type esis of malignant melanoma. and farnesylation deficient LKB1 (ref. 51). Notably, several In contrast to previous studies investigating the role of C-terminally truncated hLKB1 variants (due to nucleotide LKB1 farnesylation in the oocyte of Drosophila22, we found deletions or point mutations in STK11) with an intact kinase farnesylation of LKB1 to be dispensable for the correct have been reported52 and OMIM to be associated with localization in the embryonic epidermis and NBs and for the Peutz-Jeghers syndrome, which might be due to the function of function of LKB1 in Drosophila development. This observation the C-terminal region in PA-mediated activation of LKB1. is in line with previous findings from cultured mammalian cells Apart from the association of LKB1 with STRADa/Mo25, and mice. In malignant melanoma cells, farnesylation-deficient membrane binding and PA-mediated activation of LKB1 is a LKB1 suppresses colony formation at similar degree as the newly described upstream mechanism, which controls LKB1 wild-type kinase20. Furthermore, studies from knock-in mice kinase activity and function in vitro and in vivo. The regulation of show that mice homozygous for farnesylation deficient LKB1 are LKB1 function by membrane recruitment raises the question viable and do not exhibit any obvious phenotypes21. However, whether this depends primarily on membrane localization in these mice AMPK is less phosphorylated, whereas in cell or whether lipid binding itself stimulates LKB1 activity. Several

8 NATURE COMMUNICATIONS | 8:15747 | DOI: 10.1038/ncomms15747 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15747 ARTICLE

(phospho)-lipids have been demonstrated to activate or were performed in three independent experiments with n ¼ 100 in each increase the activity of various serine/threonine kinases, including experiment. For rescue experiments using the UAS/GAL4 system, we used Phosphoinositide-dependent kinase 1 (PDK1, activated by UAS::GFP-LKB1 and actin5C::GAL4 instead of lkb1::GFP-LKB1. PtdIns(3,4,5)P3) or PKCz, Raf1, and mTOR (all activated by 49,53–55 DNA and constructs. of the cDNA of wild-type LKB1 into pENTR was PA ). Fusion of GFP-LKB1D to lipid-binding LB C564A performed using standard PCR on a full length EST clone (Drosophila Genomics domains (PH(PLCd) and PH(Akt1)) resulted in a re-activation Resources Center, DGRC) as template using the following primers: LKB1-F: of GFP-LKB1DLB C564A (Fig. 1o; Supplementary Fig. 4). Further- 50- CACCATGCAATGTTCTAGCTCTCGG-30, LKB1-R: 50-CTACGAAGTTCG- 0 more, the activity of LKB1DLB C564A is strongly reduced in in vitro GCAGTGG-3 . Similar, truncated fragments of LKB1 were cloned with the fol- 0 0 lowing oligonucleotides: LKB1512-F: 5 -CACCATGCACACCTACGAACCGCC-3 , kinase assays (Fig. 3f,g). These findings could substantiate a 0 0 0 LKB1536-F: 5 -CACCATGGCGCCCGTCAAGAAG-3 ,LKB1552 -F: 5 -CACCAT- hypothesis, in which the membrane functions as a scaffold for GCTGACGTCCTGCATCTCCG-30. For expression of LKB1 from its endogenous bringing together LKB1 and its cofactors (STRADa and Mo25) or promoter we inserted a genomic fragment (from 2.8 kbp upstream of the its substrates. However, membrane binding deficient LKB1 translation start to 1 kbp downstream of the stop codon) into pENTR using the 0 0 following primers: LKB1gen-F: 5 - CACC CACTAGCGTAATTTGACGG-3 , robustly associates with its cofactors and one of its substrates 0 0 (AMPK, Supplementary Fig. 3b). Moreover, addition of PA LKB1gen-R: 5 - CTC GAG CAGCAGTACGGTCATCTC-3 . An XbaI-cutting site was introduced replacing the start codon using mutagenesis PCR and the following 0 (but not of PtdIns(4,5)P2 or PtdIns(3,4,5)P3) is capable to primer: LKB1gen-XbaI-F: 5 -GGCTCCGCGGAGGTTTTCTAGACAATGTTCTA- substantially enhance the kinase activity of LKB1 in vitro (Fig. 3g; GCTCTC-30. Subsequently, GFP was inserted into the XbaI site by PCR and Supplementary Fig. 3c,d). Thus it is more likely, that in addition standard ligation. Mutagenesis PCR was used to generate defined point mutations to membrane recruitment, PA functions to induce a conforma- with full length or genomic LKB1 cDNA in pENTR as template. The following oligonucleotides were used for mutagenesis (mutation underlined): LKB1DLB: tional change of LKB1 thus enhancing its kinase activity. 0 Combination of 1. LKB1K546A R547A R548A K550A K551A-F: 5 -TCGGCACTG- 0 However we cannot exclude that in our in vitro kinase assays GCGGCGGCCGCCGCGGCGCTGACGTCCTGC-3 and 2.LKB1K539A K540A 0 0 using immunoprecipiptated proteins from transfected cells wild- K541A-F: 5 -GAGGAGGCGCCCGTCGCCGCGGCGGGATCGGCACTG-3 , 0 LKB1C564A-F: 5 -GTGCGCAAGCTTAGCCACGCCCGAACTTCGTAG, type LKB1 might have co-immunoprecipitated with PA-enriched 0 micelles, which enhanced its kinase activity in this experiment. LKB1D317A-F: 5 - CAAACGCTGAAGATTTCCGCCTTCGGTGTGGCG. To express LKB1DLBC564A PH(PLD) and LKB1DLBC564A PH(Akt), the PH domain of Interestingly, PLD-produced PA activates both, LKB1 (this study) PLCd and Akt was amplified by PCR and ligated via an endogenous Bpu1101-I site and mTOR48–50,55. Up to now, overexpression of PLD and into lkb1::LKB1 pENTR using the following primers: PH(PLCd)-F: 50- GCTG- increased levels of PA have been only assigned to pro-oncogenic AGCCACGCCCGAACT TCGgatgaggatctacaggcgct-30, PH(PLCd)-R: 50- GCTGA- 54–58 GCTAGATCTTGTGCAGCCCCAG-30, PH(Akt)-F: 50-GCTGAGCCACGCCCG- pathways, in particular mTOR and Raf signalling .Ourdata AACTTCGGTCGTAAAGGAGGGGTGG-30 and PH(Akt)-R: 50- GCTGAGCTT- thus demonstrate the first anti-oncogenic mechanism of PLD ATATGAGCCGGCTGGATAC-30. function, which might counterbalance its pro-oncogenic effects as For expression of hLKB1, the open reading frame of hLKB1 was cloned into pENTR using the following nucleotides: hLKB1-F: 50- CACC ATGGAGGTG- long as the LKB1-AMPK signalling axis is intact (Fig. 7). 0 0 0 Notably, Mukhopadhyay et al. showed recently, that in a GTGGACCC-3 and hLKB1-R: 5 - TCACTGCTGCTTGCAGG-3 . For mutation of the farnesylation, lipid binding motif and construction of the kinase dead version, mesenchymal-like breast line (MDA-MB-231), which the following nucleotides were used in mutagenesis PCRs: hLKB1C430A-F: expresses robust levels of endogenous LKB1 (in contrast to the 50- CGCCGGCTGTCGGCCGCTAAGCAGCAGTGAAAGGGT-30, 0 majority of metastases), activated PLD induces a hLKB1R415AK416A-F: 5 - GCCCCCAACCCTGCCGCCGCGGCCTGCTCCGCC- 0 0 reduction in phospho-AMPK59. Thus, PLD-mediated activation AGC-3 and hLKB1D194A:5- ACCCTCAAAATCTCCGCCCTTGGCGTGGCCG- AGGCA-30. of LKB1 might be either cell type specific or depends on the basal Constructs were recloned into GFP-tagged destination vectors containing a activation of both signalling cascades, the LKB1-AMPK axis and One-Strep-Tag fused to the N terminus of GFP (USGW, modified TGW, the PI3K-AKT pathway. Murphy lab, DGRC, expression in S2R cells) or into a modified EGFP-C1 vector The LKB1-AMPK signalling pathway is well described to play a (CGW, expression in mammalian cells) containing a gateway cassette using the crucial role for cell polarity, energy homeostasis and cell proliferation gateway technology (Life technology). and is frequently disturbed in various types of cancer16,17.As Antibodies. Antisera directed against full length LKB1 were raised by injection of mentioned above, one particular function of AMPK is to a fusion protein of LKB1 and MBP into two guinea pigs (Amsbio, Abingdon, UK). counterbalance the pro-proliferative activity of mTOR8–15. Defects in the LKB1-AMPK signalling pathway (resulting in impaired Cell culture and cell viability assay. HeLa,IMR90,IGR37andMDCKcellswere inhibition of mTOR) have been already demonstrated to contribute obtained from ATCC. All cells were maintained in Dulbecco’s modified Eagle’s to BRAF- and KRAS-induced melanoma formation in mouse medium supplemented with 10% fetal calf serum, 2 mM Glutamine at 37 °Cina 60–63 models . Remarkably, somatic mutations in human STK11 5% CO2 atmosphere and negatively tested for mycoplasma contamination by PCR. (encoding LKB1) are rather rare in malignant melanoma64,65. Cells were transfected using FUGENE (Promega) according to the manufacturer’s instructions. For evaluation of cell viability upon energetic stress, 50 103 cells/well However, our data demonstrate that downregulation of LKB1 were transiently transfected in a 24well plate with GFP-hLKB1 þ STRADa constructs protein expression can be observed in a majority of malignant and treated for 12 h with 2.5 mM AICAR (Santa Cruz Inc.). GFP þ STRADa were melanoma samples and therefore most likely contributes to the used as negative control. Cell viability was assessed in triplicates using MTT assay pathogenesis of this type of skin cancer. according to the manufacturer’s instructions (SIGMA). Our study provides strong evidences for an important and conserved regulatory mechanism, which facilitates membrane Western blotting and coimmunoprecipitation. Western blotting was done as 66 recruitment of LKB1 by PA and is essential for the kinase previously described . For mammalian cell culture experiments, HeLa or IGR37 cells were transiently transfected with the indicated constructs. 48 hours after activity and function of LKB1 during development and tumour transfection, cells were incubated with 2 mM AICAR for 1 h and harvested in suppression. The dissection of upstream regulatory mechanisms lysis buffer (1% Triton X-100, 150 mM NaCl, 1 mM CaCl2, 1 mM MgCl2,50mM of LKB1 contributes to a better understanding of the physiolo- TRIS-HCl pH 7.5) supplemented with protease- and inhibitors. For gical regulation of LKB1 and their misregulation in tumours, embryonic lysates, lkb1::GFP-LKB1 expressing embryos were collected from overnight plates. Coimmunoprecipitation using GFP-Trap (ChromoTek) of which might be useful for the development of therapeutic GFP-LKB1 with HA-tagged DmSTRADa (Stlk) and myc-tagged DmMo25 was approaches in the future. done in embryonic lysates, which ubiquitously expressed the proteins using arm::GAL4. Primary antibodies used for western blotting were as follows: rabbit anti-Actin (1:1,000, sc-47778, Santa Cruz), mouse anti-S6K (1:500, sc-8418, Methods Santa Cruz ), rabbit phosphoT389-S6K (1:500, sc-11759, Santa Cruz), rabbit Fly stocks and genetics. UASt::GFP-LKB1 and lkb1::GFP-LKB1 transgenes were anti-phospho-T172-AMPK (1:200, sc-33524, Santa Cruz), rabbit anti-phospho- generated using phiC31-mediated germ line transformation on attp40. For rescue MARK1/2/3 (1:500, PA5-17495, Thermo Scientific), mouse anti-myc (1:100, 9E10, experiments of different LKB1 variants, we used the lkb1x5 null allele. Lethality tests DSHB), rabbit anti-AMPK (1:400, sc-25792, Santa Cruz), guinea pig anti LKB1

NATURE COMMUNICATIONS | 8:15747 | DOI: 10.1038/ncomms15747 | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15747

(1:500, this study), mouse anti-GFP (1:500, sc-9996, Santa Cruz), mouse anti-HA 0.3 mCi[g-32ATP] in kinase buffer for 1 h at 30 °C. The reaction was terminated (1:500, #11583816001, Roche), rabbit anti-GST (1:5,000, #G7781, SIGMA). For by addition of SDS sample buffer and samples were subjected to SDS-PAGE. statistical analysis, three independent experiments were scored. Intensity of the Phosphorylation was detected by exposure to X-ray films and quantified with bands was quantified by ImageJ. Aida 2D Densitometry software. For assays with human recombinant kinase complex (Supplementary Fig. 3b,c), 1 mg of a complex of recombinant hLKB1, STRADa and Mo25 (SIGMA) Immunohistochemistry. Drosophila embryos were fixed in 4% formaldehyde, was used as described above replacing immunoprecipitated LKB1 protein. In this 67 phosphate buffer pH 7.4 as described before . Primary antibodies used for indirect experiment, GFP-AMPKa1 from transfected HeLa cells (2 mg total protein lysate) immunofluorescence were as follows: guinea pig anti LKB1 (1:250, this study), was used as substrate as a recombinant fragment would have interfered with the rabbit anti Baz (1:2,000, kindly provided by A. Wodarz), mouse anti a-spectrin autophosphorylation bands. After isolation of precipitated OneStrep-GFP-AMPK, (3A9, 1:50, DSHB), mouse anti Dlg (4F10, 1:50, DSHB), rat anti DE-Cad recombinant hLKB1/STRADa/Mo25 was added and incubated as described above. (DCAD2, 1:25, DSHB), rabbit anti GFP (1:400, sc-8334, Santa Cruz Inc), Subsequently, OneStrep-GFP-AMPK was purified from the reaction mixture guinea-pig anti Miranda (1:1,000, kindly provided by A. Wodarz). HeLa cells using Streptactin beads. were fixed with 4% PFA in PBS and stained with the rabbit anti-phospho-Sad For kinase assays with addition of lipids, Liposomes were prepared as described antibody (1:250) and a mouse anti-LKB1 antibody (1:200, sc-32245, Santa Cruz) above and added at 10 mM final concentration. in 10% goat serum. For statistical analysis, three independent experiments were scored. Intensity of Secondary antibodies conjugated with Alexa 488, Alexa 568 and Alexa 647 the bands was quantified by ImageJ. (Life Technology) were used at 1:400. Images were taken on a Zeiss LSM 710 Meta confocal microscope and processed using Adobe Photoshop. For immunohistochemical staining of healthy skin, nevi and melanoma, a tissue Transfection and analysis of neurons. Hippocampal neurons were isolated from micro array (TMA) of paraffin-embedded healthy skin, nevi and melanoma the brains of E18 rat embryos and cultured as described previously69. Dissociated primary tumours was analysed. Paraffin sections were deparaffinized for 30 min at hippocampal neurons were plated at 70,000 cells per well in a 24 well plate and 72 °C, washed two times for 7 min in Xylol and subsequently re-watered in a transfected 3 h after plating by phosphate co-precipitation. Neurons were descending sequence of /water mixture. Prior to staining sections were fixed at 3 DIV and permeabilized with 0.1% Triton X-100, 0.1% sodium citrate in subjected for 5 min to heat-induced epitope-retrieval (HIER) using 1 mM PBS for 3 min on ice. Neurons were stained with the Tau-1 antibody as axonal Tris-EDTA-buffer (pH 8.5) at 120 °C. Sections were blocked in peroxidase- marker (Chemicon, MAB3420; 1:300) and Alexa-Fluor-conjugated secondary blocking solution (Dako, #S2023) for 5 min at room-temperature and washed antibodies (Molecular Probes; 1:300). The stage of neuronal differentiation and 5 min with wash buffer (Dako, #S3006) prior to incubation with primary antibody axon formation was determined according to published criteria70. For each LKB1 diluted in antibody diluent (Dako, #S2022) for 30 min. Primary antibodies were as variant, 3 different experiments with 100 analysed neurons in each were scored. follows: rabbit anti-LKB1 (D60C5F10, 1:200, ), rabbit anti-phospho- Akt (1:20, Cell Signaling #4060), rabbit anti-PLD2 (1:1,000, Cell Signaling #13891) and rabbit anti-S6K-phospho-T389 S6K (1:50, Cell Signaling #9206). Subsequently, Statistics. All experiments were performed in triplicates. Error bars represent s.d. sections were washed with wash buffer and incubated for 30 min with and statistical significance was determined using ANOVA: Po0.0001, HRP-coupled secondary antibody (Dako EnVision, # K5007). Stainings were ****Po0.001, ***Po0.01, **Po0.05, *P40.05, not significant (NS). developed after washing using DAB/chromogen solution (Dako, #K5007). To visualize cellular structures, stained sections were counterstained with hematoxylin Data availability. The authors declare that all data are available within the Article (Merk, #10517505000) for 1 min and dehydrated in ethanol/xylol before and Supplementary Files, or available from the authors upon request. embedding. The staining intensity was determined blinded for all tissue samples as followed: negative-0, weak- þ , moderate- þþ and strong- þþþ.LKB1 exhibited a strong expression in melanocytes in situ, so weak or negative intensity References was scored as downregulation of the protein, whereas moderate staining was 1. Lizcano, J. M. et al. LKB1 is a master kinase that activates 13 kinases of the classified as slightly downregulated. Activated Akt was negative in healthy skin AMPK subfamily, including MARK/PAR-1. EMBO J. 23, 833–843 (2004). biopsies, so weak, moderate and strong staining was classified as upregulated. PLD2 2. Vaahtomeri, K. & Makela, T. P. Molecular mechanisms of tumor suppression expression was negative or weak in melanocytes in situ, thus moderate and strong by LKB1. FEBS Lett. 585, 944–951 (2011). staining was scored as upregulated. Phospho-S6K was strongly expressed in the 3. Shaw, R. J. et al. The tumor suppressor LKB1 kinase directly activates nucleus but not in the cytoplasm of melanocytes but cytoplasmic staining occurred AMP-activated kinase and regulates apoptosis in response to energy stress. only in malignant tumours. Consequently, weak to strong cytoplasmic staining of Proc. Natl Acad. Sci. USA 101, 3329–3335 (2004). phospho-S6K was classified as upregulated. 4. Lee, J. H. et al. Energy-dependent regulation of cell structure by AMP-activated . Nature 447, 1017–1020 (2007). Lipid binding assays. Fusion proteins of the C terminus of LKB1 (aa 353–567) 5. van der Velden, Y. U. et al. The serine-threonine kinase LKB1 is essential for with MBP were expressed in E.coli and affinity purified. Lipid strips (Echelon) were survival under energetic stress in zebrafish. Proc. Natl Acad. Sci. USA 108, incubated over night with purified MBP-LKB1 fusion proteins at 0.5 mgml 1 in 4358–4363 (2011). TBST containing 3% BSA, washed and probed with antibodies against MBP 6. Hawley, S. A. et al. Complexes between the LKB1 tumor suppressor, STRAD (1:1,000, Santa Cruz sc-73416) as described above. alpha/beta and MO25 alpha/beta are upstream kinases in the AMP-activated For membrane floatation and in vitro kinase assays, lipids were obtained from protein kinase cascade. J. Biol. 2, 28 (2003). Avanti Polar Lipids (Egg-PA, #840101, Egg-PC #840051, Brain PtdIns(4,5)P2 7. Woods, A. et al. LKB1 is the upstream kinase in the AMP-activated protein #840046) and Echelon (PtdIns(3,4,5)P3 #P-3916). Liposomes (10 mM total lipid kinase cascade. Curr. Biol. 13, 2004–2008 (2003). concentration, either PC alone or PC:PA/PtdIns(3,4,5)P3/PtdIns(4,5)P2 in a 9:1 8. Inoki, K., Zhu, T. & Guan, K. L. TSC2 mediates cellular energy response to molar ratio) were prepared in LB buffer (30 mM Tris, 4 mM EGTA, pH 8.0) by control cell growth and survival. Cell 115, 577–590 (2003). extrusion through a 0.1 mm polycarbonate membrane using a Mini-Extruder 9. Shaw, R. J. et al. The LKB1 tumor suppressor negatively regulates mTOR (Avanti Polar Lipids). The membrane floatation was performed as follows68: signaling. Cancer Cell 6, 91–99 (2004). liposomes (100 ml of 10 mM total lipid concentration) were incubated on ice for 10. Gwinn, D. M. et al. AMPK phosphorylation of raptor mediates a metabolic 30 min with 1 mg recombinant protein. LB buffer (30 mM Tris, 4 mM EGTA, checkpoint. Mol. Cell 30, 214–226 (2008). 2 M sucrose (pH 8.0)) was added to the incubation reaction to bring the final 11. Arsham, A. M., Howell, J. J. & Simon, M. C. A novel hypoxia-inducible sucrose concentration to 1.6 M, and this mixture was overlaid with cushions factor-independent hypoxic response regulating mammalian target of containing 1.4 M, 0.4 M, and 0.25 M sucrose in the same buffer in a TLA-55 tube. rapamycin and its targets. J. Biol. Chem. 278, 29655–29660 (2003). After centrifugation at 186,000g (4 °C) for 45 min in a TLA-55 rotor (Beckman), 12. Kimura, N. et al. A possible linkage between AMP-activated protein kinase the 0.25/0.4 M interphase (top fraction, T) and the loading fraction (bottom (AMPK) and mammalian target of rapamycin (mTOR) signalling pathway. fraction, B) were collected and analysed by SDS-PAGE and western blot. Cells 8, 65–79 (2003). For statistical analysis, three independent experiments were scored. Intensity 13. Shackelford, D. B. et al. mTOR and HIF-1alpha-mediated tumor in of the bands was quantified by ImageJ. an LKB1 mouse model of Peutz-Jeghers syndrome. Proc. Natl Acad. Sci. USA 106, 11137–11142 (2009). In vitro kinase assay. OneStrep-GFP-LKB1 plus OneStrep-GFP-Stlk were 14. Corradetti, M. N., Inoki, K., Bardeesy, N., DePinho, R. A. & Guan, K. L. precipitated from transfected S2R cells (2 mg total protein lysates) using Streptactin Regulation of the TSC pathway by LKB1: evidence of a molecular link beads (IBA, Goettingen, Germany). The beads were washed five times in harsh between complex and Peutz-Jeghers syndrome. Genes Dev. washing buffer (50 mM TRIS pH 7.5, 500 mM NaCl) and one time in LKB1 18, 1533–1538 (2004). kinase buffer (50 mM TRIS pH 7.5, 10 mM MgCl2, 10 mM MnCl2, 1 mM DTT, 15. Shaw, R. J. et al. The kinase LKB1 mediates glucose homeostasis in liver and 100 mM ATP, phosphatase- and protease inhibitors). Immunoprecipitated therapeutic effects of metformin. Science 310, 1642–1646 (2005). proteins were then incubated with 2 mg of recombinant GST-DmAMPKa108-280 16. Hardie, D. G. & Alessi, D. R. LKB1 and AMPK and the cancer-metabolism ( ¼ aa 108–280, containing the T-loop with the LKB1-phosphorylation site) and link—ten years after. BMC Biol. 11, 36 (2013).

10 NATURE COMMUNICATIONS | 8:15747 | DOI: 10.1038/ncomms15747 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15747 ARTICLE

17. Nakano, A. & Takashima, S. LKB1 and AMP-activated protein kinase: 43. Barnes, A. P. et al. LKB1 and SAD kinases define a pathway required for the regulators of cell polarity. Genes Cells 17, 737–747 (2012). polarization of cortical neurons. Cell 129, 549–563 (2007). 18. Boudeau, J. et al. MO25alpha/beta interact with STRADalpha/beta enhancing 44. Shelly, M., Cancedda, L., Heilshorn, S., Sumbre, G. & Poo, M. M. LKB1/STRAD their ability to bind, activate and localize LKB1 in the cytoplasm. EMBO J. 22, promotes axon initiation during neuronal polarization. Cell 129, 565–577 5102–5114 (2003). (2007). 19. Dorfman, J. & Macara, I. G. STRADalpha regulates LKB1 localization by 45. Veleva-Rotse, B. O. et al. STRAD pseudokinases regulate axogenesis and LKB1 blocking access to importin-alpha, and by association with Crm1 and stability. Neural Dev. 9, 5 (2014). exportin-7. Mol. Biol. Cell. 19, 1614–1626 (2008). 46. Saito, M. et al. Expression of phospholipase D2 in human colorectal carcinoma. 20. Sapkota, G. P. et al. Phosphorylation of the protein kinase mutated in Oncol. Rep. 18, 1329–1334 (2007). Peutz-Jeghers cancer syndrome, LKB1/STK11, at Ser431 by p90(RSK) and 47. Henkels, K. M., Boivin, G. P., Dudley, E. S., Berberich, S. J. cAMP-dependent protein kinase, but not its farnesylation at Cys(433), is & Gomez-Cambronero, J. Phospholipase D (PLD) drives cell invasion, tumor essential for LKB1 to suppress cell vrowth. J. Biol. Chem. 276, 19469–19482 growth and metastasis in a human breast cancer xenograph model. (2001). 32, 5551–5562 (2013). 21. Houde, V. P. et al. Investigation of LKB1 Ser431 phosphorylation and Cys433 48. Yoon, M. S. et al. Rapid mitogenic regulation of the mTORC1 inhibitor, farnesylation using mouse knockin analysis reveals an unexpected role of DEPTOR, by phosphatidic acid. Mol. Cell. 58, 549–556 (2015). prenylation in regulating AMPK activity. Biochem. J. 458, 41–56 (2014). 49. Fang, Y., Vilella-Bach, M., Bachmann, R., Flanigan, A. & Chen, J. Phosphatidic 22. Martin, S. G. & St Johnston, D. A role for Drosophila LKB1 in anterior- acid-mediated mitogenic activation of mTOR signaling. Science 294, 1942–1945 posterior axis formation and epithelial polarity. Nature 421, 379–384 (2003). (2001). 23. Song, P. et al. Protein kinase Czeta-dependent LKB1 serine 428 50. Foster, D. A., Salloum, D., Menon, D. & Frias, M. A. Phospholipase D and the phosphorylation increases LKB1 nucleus export and apoptosis in endothelial maintenance of phosphatidic acid levels for regulation of mammalian target of cells. J. Biol. Chem. 283, 12446–12455 (2008). rapamycin (mTOR). J. Biol. Chem. 289, 22583–22588 (2014). 24. Tiainen, M., Ylikorkala, A. & Makela, T. P. Growth suppression by Lkb1 is 51. Denison, F. C., Hiscock, N. J., Carling, D. & Woods, A. Characterization of an mediated by a G(1) cell cycle arrest. Proc. Natl Acad. Sci. USA 96, 9248–9251 alternative splice variant of LKB1. J. Biol. Chem. 284, 67–76 (2009). (1999). 52. Boudeau, J., Sapkota, G. & Alessi, D. R. LKB1, a protein kinase regulating cell 25. Smith, D. P., Spicer, J., Smith, A., Swift, S. & Ashworth, A. The mouse proliferation and polarity. FEBS Lett. 546, 159–165 (2003). Peutz-Jeghers syndrome Lkb1 encodes a nuclear protein kinase. 53. Limatola, C., Schaap, D., Moolenaar, W. H. & van Blitterswijk, W. J. Hum. Mol. Genet. 8, 1479–1485 (1999). Phosphatidic acid activation of -zeta overexpressed in COS 26. Baas, A. F. et al. Activation of the tumour suppressor kinase LKB1 by the cells: comparison with other protein kinase C isotypes and other acidic lipids. STE20-like pseudokinase STRAD. EMBO J. 22, 3062–3072 (2003). Biochem. J. 304(Pt 3): 1001–1008 (1994). 27. Sebbagh, M., Santoni, M. J., Hall, B., Borg, J. P. & Schwartz, M. A. Regulation of 54. Ghosh, S., Strum, J. C., Sciorra, V. A., Daniel, L. & Bell, R. M. Raf-1 kinase LKB1/STRAD localization and function by E-cadherin. Curr. Biol. 19, 37–42 possesses distinct binding domains for phosphatidylserine and phosphatidic (2009). acid. Phosphatidic acid regulates the translocation of Raf-1 in 12-O- 28. Bonaccorsi, S. et al. The Drosophila Lkb1 kinase is required for spindle tetradecanoylphorbol-13-acetate-stimulated Madin-Darby canine kidney cells. formation and asymmetric neuroblast division. Development 134, 2183–2193 J. Biol. Chem. 271, 8472–8480 (1996). (2007). 55. Toschi, A. et al. Regulation of mTORC1 and mTORC2 complex assembly by 29. Watts, J. L., Morton, D. G., Bestman, J. & Kemphues, K. J. The C. elegans par-4 phosphatidic acid: competition with rapamycin. Mol. Cell. Biol. 29, 1411–1420 gene encodes a putative serine-threonine kinase required for establishing (2009). embryonic asymmetry. Development 127, 1467–1475 (2000). 56. Zheng, Y. et al. Phospholipase D couples survival and migration signals 30. Collins, S. P., Reoma, J. L., Gamm, D. M. & Uhler, M. D. LKB1, a novel in stress response of human cancer cells. J. Biol. Chem. 281, 15862–15868 serine/threonine protein kinase and potential tumour suppressor, is (2006). phosphorylated by cAMP-dependent protein kinase (PKA) and prenylated 57. Shi, M., Zheng, Y., Garcia, A., Xu, L. & Foster, D. A. Phospholipase D provides in vivo. Biochem. J. 345(Pt 3): 673–680 (2000). a survival signal in human cancer cells with activated H-Ras or K-Ras. 31. McTaggart, S. J. Isoprenylated proteins. Cell. Mol. Life Sci. 63, 255–267 (2006). Cancer. Lett. 258, 268–275 (2007). 32. Stace, C. L. & Ktistakis, N. T. Phosphatidic acid- and phosphatidylserine- 58. Rizzo, M. A. et al. Phospholipase D and its product, phosphatidic acid, mediate binding proteins. Biochim. Biophys. Acta 1761, 913–926 (2006). agonist-dependent raf-1 translocation to the plasma membrane and the 33. Varnai, P. & Balla, T. Visualization of phosphoinositides that bind pleckstrin activation of the mitogen-activated protein kinase pathway. J. Biol. Chem. 274, domains: calcium- and agonist-induced dynamic changes and 1131–1139 (1999). relationship to myo-[3H]inositol-labeled phosphoinositide pools. J. Cell Biol. 59. Mukhopadhyay, S. et al. Reciprocal regulation of AMP-activated protein kinase 143, 501–510 (1998). and phospholipase D. J. Biol. Chem. 290, 6986–6993 (2015). 34. Pawelczyk, T. & Matecki, A. -delta3 binds with high specificity 60. Damsky, W. et al. mTORC1 activation blocks BrafV600E-induced growth to 4,5-bisphosphate and phosphatidic acid in bilayer arrest but is insufficient for melanoma formation. Cancer Cell 27, 41–56 (2015). membranes. Eur. J. Biochem. 262, 291–298 (1999). 61. Liu, W. et al. LKB1/STK11 inactivation leads to expansion of a prometastatic 35. Klippel, A., Kavanaugh, W. M., Pot, D. & Williams, L. T. A specific product of tumor subpopulation in melanoma. Cancer Cell 21, 751–764 (2012). phosphatidylinositol 3-kinase directly activates the protein kinase Akt through 62. Zheng, B. et al. Oncogenic B-RAF negatively regulates the tumor suppressor its pleckstrin homology domain. Mol. Cell. Biol. 17, 338–344 (1997). LKB1 to promote melanoma cell proliferation. Mol. Cell. 33, 237–247 (2009). 36. Bruntz, R. C., Taylor, H. E., Lindsley, C. W. & Brown, H. A. Phospholipase D2 63. Esteve-Puig, R., Canals, F., Colome, N., Merlino, G. & Recio, J. A. Uncoupling mediates survival signaling through direct regulation of Akt in glioblastoma of the LKB1-AMPKalpha energy sensor pathway by growth factors and cells. J. Biol. Chem. 289, 600–616 (2014). oncogenic BRAF. PLoS ONE 4, e4771 (2009). 37. Menon, S. & Manning, B. D. Common corruption of the mTOR signaling 64. Guldberg, P. et al. Somatic mutation of the Peutz-Jeghers syndrome gene, network in human tumors. Oncogene 27, S43–S51 (2008). LKB1/STK11, in malignant melanoma. Oncogene 18, 1777–1780 (1999). 38. Gomez-Cambronero, J. Phospholipase D in cell signaling: from a myriad of cell 65. Rowan, A. et al. Somatic mutations in the Peutz-Jeghers (LKB1/STKII) functions to cancer growth and metastasis. J. Biol. Chem. 289, 22557–22566 gene in sporadic malignant melanomas. J. Invest. Dermatol. 112, 509–511 (2014). (1999). 39. Henkels, K. M., Short, S., Peng, H. J., Di Fulvio, M. & Gomez-Cambronero, J. 66. Cidlinsky, N. et al. Inactivation of the LKB1-AMPK signaling pathway does not PLD2 has both enzymatic and cell proliferation-inducing capabilities, that are contribute to salivary gland tumor development—a short report. Cell Oncol. 39, differentially regulated by phosphorylation and dephosphorylation. Biochem. 389–396 (2016). Biophys. Res. Commun. 389, 224–228 (2009). 67. Sen, A., Nagy-Zsver-Vadas, Z. & Krahn, M. P. Drosophila PATJ supports 40. LaLonde, M. M. et al. Regulation of phototransduction responsiveness and adherens junction stability by modulating Myosin light chain activity. J. Cell retinal degeneration by a phospholipase D-generated signaling lipid. J. Cell Biol. Biol. 199, 685–698 (2012). 169, 471–479 (2005). 68. Krahn, M. P., Klopfenstein, D. R., Fischer, N. & Wodarz, A. Membrane 41. Sato, T., Hongu, T., Sakamoto, M., Funakoshi, Y. & Kanaho, Y. Molecular targeting of Bazooka/PAR-3 is mediated by direct binding to phosphoinositide mechanisms of N-formyl-methionyl-leucyl-phenylalanine-induced superoxide lipids. Curr. Biol. 20, 636–642 (2010). generation and degranulation in mouse neutrophils: phospholipase D is 69. Yang, R., Kong, E., Jin, J., Hergovich, A. & Puschel, A. W. Rassf5 and Ndr dispensable. Mol. Cell. Biol. 33, 136–145 (2013). kinases regulate neuronal polarity through Par3 phosphorylation in a novel 42. Raghu, P., Manifava, M., Coadwell, J. & Ktistakis, N. T. Emerging findings pathway. J. Cell. Sci. 127, 3463–3476 (2014). from studies of phospholipase D in model organisms (and a short update 70. Schwamborn, J. C. & Puschel, A. W. The sequential activity of the GTPases on phosphatidic acid effectors). Biochim. Biophys. Acta 1791, 889–897 Rap1B and Cdc42 determines neuronal polarity. Nat. Neurosci. 7, 923–929 ð2009Þ: (2004).

NATURE COMMUNICATIONS | 8:15747 | DOI: 10.1038/ncomms15747 | www.nature.com/naturecommunications 11 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15747

Acknowledgements How to cite this article: Dogliotti, G. et al. Membrane-binding and activation of We thank A. Wodarz, B. Schmidt, G. La¨ngst, the Bloomington Drosophila stock center at LKB1 by phosphatidic acid is essential for development and tumour suppression. the University of Indiana and the Developmental Studies Hybridoma Bank at the Nat. Commun. 8, 15747 doi: 10.1038/ncomms15747 (2017). University of Iowa for providing reagents. This work was supported by grants of the Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in DFG to M.P.K. (DFG3901/2-1, DFG3901/1-2, SFB699/A13) and to A.W.P. (DFG PU 102 published maps and institutional affiliations. 12-1 and Cells-in-Motion Cluster of Excellence (EXC 1003—CiM)).

Author contributions Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, G.D., L.K., C.T., P.D., O.P. and G.M. designed and performed the experiments, A.W.P. adaptation, distribution and reproduction in any medium or format, as long as you give and M.P.K. designed and directed the project and wrote the manuscript. All authors appropriate credit to the original author(s) and the source, provide a link to the Creative discussed results and commented on the manuscript. Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the Additional information article’s Creative Commons license and your intended use is not permitted by statutory Supplementary Information accompanies this paper at http://www.nature.com/ regulation or exceeds the permitted use, you will need to obtain permission directly from naturecommunications the copyright holder. To view a copy of this license, visit http://creativecommons.org/ Competing interests: The authors declare no competing financial interests. licenses/by/4.0/

Reprints and permission information is available online at http://npg.nature.com/ reprintsandpermissions/ r The Author(s) 2017

12 NATURE COMMUNICATIONS | 8:15747 | DOI: 10.1038/ncomms15747 | www.nature.com/naturecommunications